Differential Equations.
Equations. Partial Differential Equation
∂2 f ∂2 f ∂2 f ∂f ∂f
A + B + C + D + E + Ff + G = 0
∂x 2
∂x∂y ∂y 2
∂x ∂y
Parabolic Equation (Diffusion Eq.)
Hyperbolic Equation (Advection Eq.) Elliptic B 2 − 4 AC < 0
p Equation
Elliptic q ((Poisson Eq.)
q) Parabolic B 2 − 4 AC = 0
∂2 f ∂2 f
+ = ρ = const (x ∈ S)
Hyperbolic
H b li Equation
E ti ∂x 2 ∂y 2
∂ f
2
∂ f
2
Wave Eq. − c2 2 = 0 Boundary Conditions : f is given at the surface - S
(A = −c2,B = 0,C = 1)
∂t 2
∂x
3 4
Partial Differential Equation
Differential Equation in time
in time and space
・Initial Condition t a Boundary
・Initial ou da y Problem
ob e
Time goes in one way
One-dimensional diffusion equation:
2
d r
Newton Eq. : =F (t0 ≤ t ≤ t1 ) ∂f ∂2 f
dt 2 =κ 2 (0 ≤ x ≤ 1)
Initial Condition: r (t0 ) = r0 , r&(t0 ) = r&0 ∂t ∂x
Initial Condition: f ( x, 0) = f 0 ( x ) (0 ≤ x ≤ 1)
dv dr
: =F, = v (t0 ≤ t ≤ t1 )
dt dt Boundary Condition: f (0, t ) = 0 , f (1, t ) = 0
Initial Condition: r (t0 ) = r0 , v (t0 ) = v0
5 6
f i −n1+1 f i n +1 f i +n1+1 f i n +1
t =t n +1 n +1
t
f i −n1 fi n f i +n1
Δt n
f i −n1 fi n f i +n1 t
i−1 i i+1
t =t n
n −1
x1 = 0 xi −1 xi xi +1 xN = 1
t
Δx
n +1
f ( xi , t n ) = f i f ( xi + Δx, t n + Δt ) = f i +1
n xi −1 xi xi +1
7 8
Introduction to finite
Difference Method Forward Difference
∂f f − fi ∂f
→ i +1 =
∂x x=xi
+ O(Δx) (1)
∂x Δx
The accuracy of the finite difference is the first order of Δx .
9 10
∂f 1 ∂2 f 1 ∂3 f ∂f 1 ∂3 f
f ( xi − Δx) = f ( xi )− Δx + Δx − 2
Δx + ⋅ ⋅ ⋅
3 f ( xi + Δx) = f ( xi − Δx) + 2 Δx + Δx3 + ⋅ ⋅ ⋅
∂x x = xi 2 ∂x 2 x = xi
6 ∂x 3 x = xi
∂x x = xi 3 ∂x3 x = xi
f i − f i −1 ∂f 1 ∂2 f 1 ∂3 f f i +1 − f i −1 ∂f 1 ∂3 f
= − Δx + Δx + ⋅ ⋅ ⋅
2
= + Δx 2 + ⋅ ⋅ ⋅
Δx ∂x x = xi 2 ∂x 2 x = xi
6 ∂x 3 x = xi 2Δx ∂x x = xi 6 ∂x3 x = xi
∂f ∂f
= + O(Δx)
∂x (2) = + O(Δx 2 )
x = xi
∂x x = xi
The accuracy of the finite difference is the first order of Δx . The accuracy of the finite difference is the second order of Δx .
11 12
Accuracy of Finite Difference Typical
yp Finite Difference Expressions
p
Expressions for 1st-
1st-order derivative
f i +1 − f i
∂f f i +1 − f i
Error 10
0
Δx = (Δx )
= k x cos k x xi − F .D. −1
1 ∂x Δx
10
∂f f i +1 − f i −1
10
−2
f i +1 − f i −1
∂x
=
2Δx
(Δx )
2
2Δx
Error
−3
∂f − f i + 2 + 4 f i +1 − 3 f i
k x = 22π
π
10
= (Δx )
2
E
10
−4
∂x 2Δx
f i = sin k x xi ∂f − f i + 2 + 6 f i +1 − 3 f i − 2 f i −1
10
−5
− f i + 2 + 8 f i +1 − 8 f i −1 + f i − 2 ∂x
=
6Δx
(Δx )
3
−6
10 12Δx
∂f − f i + 2 + 8 f i +1 − 8 f i −1 + f i − 2
Source Code 10
−7
−3 −2 −1 ∂x
=
12Δx
(Δx )
4
10 10 Δx 10 10
0
13 14
Δx2 ∂x 12 ∂x4
∂ 2 f − f i + 2 + 16 f i +1 − 30 f i + 16 f i −1 − f i − 2
( )
x= xi x= xi
∂2 f = Δx 4
= + O(Δx2 ) ∂x 2
12Δx 2
∂x2 x= xi
Replacing 1/5
1 with the 2/5 2/5 2/5
average
1 1 1 2/5 1 2/5
Initial 1/5 1/5
1
2/5 2/5 2/5
1/5
17 18
G
Game Rule: get a new value by replacing
the average about surroundings Game
Game
f i , j +1
Thiss average
a e age p
process
ocess is
s
f i , j + f i +1, j + f i −1, j + f i , j +1 + f i , j −1
f i *, j =
5
19 20
Game
Game Game
Game
n +1
f i ,nj + f i +n1, j + f i −n1, j + f i ,nj +1 + f i ,nj −1 When we regard Δ t = 1.0, Δ x = Δ y = 1.0, κ = 1/5,
f =
f i ,nj+1 − f i ,nj f i +n1, j − 2 f i ,nj + f i −n1, j f i ,nj +1 − 2 f i ,nj + f i ,nj −1
i, j
5
=κ +κ
n : present value Δt Δx 2 Δy 2
n+1 : the value after
f average
g
2 di
2-dimendional
di l diff
diffusion
i equation
ti
By subtracting f i , j from both side
∂f ⎛ ∂2 f ∂2 f ⎞
f i ,nj+1 − f i ,nj =
f n
i +1, j −2f + fn n
i −1, j +f n
i , j +1 −2f + f
n n
i , j −1 = κ⎜⎜ 2 + 2 ⎟⎟
∂t ⎝ ∂x ∂y
i, j i, j
5 ⎠
21 22
1-dimensional
P b li E
Parabolic Equation
ti Diffusion Equation
∂φ ∂ φ ∂φ ∂ 2φ
=κ 2
2
1-dimeisional
diff i eq.
diffusion =κ 2 κ : diffusion coefficient
∂t ∂x κ : diffusion coefficient
∂t ∂x
Applying the forward finite difference to the time derivative term
and the center finite difference to the spatial difference,
Image of diffusion equation: spreading distribution
with fading out.
out φ nj +1 − φ nj φ nj +1 − 2φ nj + φ nj −1
=κ
→ Increasing entropy Δt Δx 2
( )
Particle Collision Process from the microscopic view
Thermal Conduction:
Conduction: Electron Collision
We can reduce to φ nj +1 = φ nj + μ φ nj +1 − 2φ nj + φ nj −1
Viscousity: Ion Collision
κΔt
Nuclear Reactor:
Reactor: Neutron Collision where μ=
23 Δx 2 24
Sample Program 1
Sample Stability Analysis (1/3)
(1/3)
#include "xwin.h"
#define N 101 von Neumann’s Method
n ik ⋅ jΔx
Assuming the perturbation φ j = δφ e
n
int main() {
double f[N], fn[N], x[N], mu = 0.25; Where the notation i is the imaginary, k is the wave
i t j,
int j icnt
i t = 0;
0 number, j is grid index, jΔx is the grid position.
(
δφnj +1 δφnj = 1 + μ e − ikΔx − 2 + eikΔx ) μ< 0 : unstable,
0 < μ < 1/2 : stable
1/2 < μ : unstable depending on the value k
= 1 − 2μ(1 − cos kΔx) We consider only the case of 0 < κ,
eikΔx + e − ikΔx 1 1 Δx 2
cos kΔx =
2 μ< Δt <
2 2 κ
δφn +1 δφn < 1 : The amplitude of the perturbation
decrease in time
time. W have
We h to
t choose
h Δt satisfying
ti f i the
th condition,
diti butt Δt should
b h ld
The calculation is stable. be decrease propotionally to Δx with decrease of Δx.
2
27 28
2-dimendional
Diffusion Equation
∂f ⎛ ∂2 f ∂2 f ⎞
= κ⎜⎜ 2 + 2 ⎟⎟
∂t ⎝ ∂x ∂y ⎠